You are on page 1of 8

Expert Systems with Applications 38 (2011) 1155211559

Contents lists available at ScienceDirect

Expert Systems with Applications


journal homepage: www.elsevier.com/locate/eswa

Using system dynamics for simulation and optimization of one coal industry
system under fuzzy environment
Jiuping Xu , Xiaofei Li
Uncertainty Decision-Making Laboratory, Sichuan University, Chengdu 610064, PR China

a r t i c l e

i n f o

Keywords:
Coal industry planning
Fuzzy set theory
System dynamics
Multiple objective programming
Simulation

a b s t r a c t
In this paper, we have developed a model that integrates system dynamics with fuzzy multiple objective
programming (SD-FMOP). This model can be used to study the complex interactions in a industry system.
In the process of conrming sensitive parameters and fuzzy variables of the SD model, we made use of
fuzzy multi-objective programming to help yield the solution. We adopted the chance-constraint programming model to convert the fuzzy variables into precise values. We use genetic algorithm to solve
FMOP model, and obtain the Pareto solution through the programming models. It is evident that FMOP
is effective in optimizing the given system to obtain the decision objectives of the SD model. The results
recorded from the SD model are in our option, reasonable and credible. These results may help governments to establish more effective policy related to the coal industry development.
 2011 Elsevier Ltd. All rights reserved.

1. Introduction
The issue about Coal industry, one important goal of which is to
extend the industrial chain reasonably and optimally, maximize
the per unit output value, has been broadly concerned and discussed all over the world (Suwala & Labys, 2002; Zhang & Zhao,
1999). Coal is not only a main energy source for many industries
and the residential sector, but also important raw material of many
chemical industry. Coal industry plays a pivotal role in social and
economic development, and will have important inuence on the
low carbon economy and the circular economy (CE) in the predictable future.
The development of the coal industry in the city we have chosen
in China plays on the directional functions. In this region the coal
industry is a complex system that includes the production of many
secondary and tertiary products. In order to research the direction
of the coal industrys development, we must understand the characteristics of the coal industry in this region. In the quantitative
analysis, quantitative values such as those pertaining to product
distribution, price and transforming rates must be obtained. These
variables are usually decided by the authority leading the industry
and are subject to many external inuences. It is evident that some
variables accord with the fuzzy standard. We introduce system
dynamics- fuzzy multiple objective programming (SD-FMOP) in
order to circumvent the uncertainties inherent to the quantitative
variables based on fuzzy variable theory. Fuzzy variable theory can
effectively describe the pricing of coal industry products. We elect
Corresponding author. Tel.: +86 28 85418522; fax: +86 28 85418522.
E-mail address: xujiuping@scu.edu.cn (J. Xu).
0957-4174/$ - see front matter  2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.03.033

the chance-constrained programming which was pioneered by


Charnes and Cooper as a means of handling uncertainty by specifying a condence level at which it is desired that the fuzzy constraint holds.
System dynamics (SD) is a simulation technology that studies
complex great systems based on the foundation of feedback control
theory and the measurement of computer imitation technology
(Min, Kyung, & Moosung, 2005). This method is a innovative subject, established in the mid-1950s by MIT professor Forrester Jay,
Nathaniel, Mass, Charles, and Ryan (1976) based on summarizing
operational research and the synthesis of system theories including
cybernetics, information feedback theory, decision-making theory,
information theory, system dynamics simulation and computer science. It is believed that SD methods provide us with tools for better
understanding complex problems (Chen et al., 2004). SD emphasizes the behavior and trends of the complex great system. The
application of simulation technology is intended to forecast the future trends according to certain parameters and levels which are
difcult to estimate. This paper introduces fuzzy programming
based on the SD (SD-FMOP) to optimize the uncertain variables
and try to obtain optimum results in simulation, and the concept
is shown in Fig. 1. We use SD to demonstrate the complex systems
of the process in which different kinds of output produced by original coal resource. A satised production value can be achieved
through the optimization of coal industry system.
The rest of the paper is structured as follows: Section 2 present
the literature reviews. Section 3 presents the general SD-FMOP
model and the solution methods. Section 4 presents the coal
industry system and discusses the use of SD-FMOP models and
applies these models to the coal industry system. Section 5

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

Fig. 1. Coal producing system.

presents the simulation and analysis. Conclusions are presented in


Section 6.
2. Literature reviews
Many researches all of the world focus on coal industry issues.
Pendharkar solve production scheduling problem in coal industry
(Pendharkar, 1997). Fan et al. discussed the investment problem
in coal industry using a system dynamics model (Fan, Yang, &
Wei, 2007). Some researches focus on coal blending optimization
under uncertain environment. Shih and Frey developed a multiobjective chance-constrained optimization model (Shih &
Christopher Frey, 1995), fuzzy set theoretic approach has been used
and a corresponding model has been developed (Chakraborty &
Chandra, 2005) by Chakraborty and Chandradealing with the optimal planning for blending raw coal of different grades. Coal supply
and balancing problem is discussed by Wojciech to select policies
that will lead to sustainability in the coal industry (Suwala, 2008).
Wojciech also concentrate on market transition and regional
adjustments in the Polish coal industry (Suwala & Labys, 2002).
An expert system in the coal mining industry called CMEOC (Zhang
& Zhao, 1999) is developed by Zhang and Zhao, the techniques of
which include multi-objective programming, fuzzy sets, and integer programming.
Many other studies include information about new products,
new technologies, and new methods of analysis. Australian Bureau
of Agriculture and Resource Economics investigated Chinas coal
reserves and the role of state-owned key mines, state-owned local
mines, and town or village owned mines, and predicted coal production and supply (Ball, Hansard, Curtotti, & Schneider, 2003).
In Japan, the effects of coal types, temperature, and the pressure
on volatilization and char gasication have been quantitatively
evaluated and analyzed under high pressure using various types
of small test furnaces (Noriaki, 2000). A zero-emission coal technology proposed by ZECA Corporation has been presented and discussed as a process that can produce electricity at 6070%
efciency with zero emissions to the atmosphere (Grzegorz,
2006). Other works have investigated the potential use of carbon-rich by-products from Integrated Coal Gasication Combined
Cycle (IGCC) and Pulverized Coal Combustion (PCC) power plants
for Hg and NOx adsorption (Aurora, Rodney, Rolando, Jack, &
Thomas, 2005). Jorjani uses articial neural networks to predict
the organic and inorganic sulfur reduction from coal using mixed
culture ferrooxidans species extracted from coal washery tailings
for pyritic sulfur and Rhodococcus species extracted from oily soils
for organic sulfur removal (Jorjani, Chelgani, & Mesroghli, 2007). It
was observed in this work that organic and inorganic sulfur reduction from coal could be satisfactorily predicted using the ANN
model. Jaffrennous studies on the environmental (Jaffrennou
et al., 2007) impact of a coal immersion accident at sea reveal that

11553

there would be limited environmental consequences produced by


this type of accident. Wigley examines combustion trials performed to determine whether mineral additives could increase
the proportion of coal ash depositing on boiler walls and whether
mineral additions could increase the y ash particle size distribution (Wigley, Williamson, & Riley, 2007). His ndings hold that
these additives have an effect on deposit growth and microstructure but do not reverse the impact of low-NOx burners on coal
ash behavior. Naill and Geinzer established a series of energy
models such as COAL1,COAL2, FOSSIL1 and FOSSIL2 based on system dynamics (Naill & Geinzer, 1993). They set up the integrated
dynamic energy analysis simulation (IDEAS) model to simulate
the American energy supply and demand system during the last
3040 years and evaluate the feasibility of energy policies. Among
the myriad publications, very few focus on coal industry planning
and related economic analysis using quantitative decision models.
3. SD-FMOP modelling
We will now describe general SD-FMOP modelling.
3.1. Modelling
In this section, we set up the universal form of the SD-FMOP
model. The general SD-FMOP modelling is shown in Fig. 2. Factors
affect each other and this is by design. This allows the system to
keep developing. By the analysis above, we observed that complex
interactions between inherent components must drive the growth
of system: steering and governance are limited by knowledge of
these interactions. To study them, we divided the system into ve
sets: initial set, scheme set, product set, fuzzy variable and objective set. We let G express the coal set, X be the scheme set, Z be
the product value set, C be the fuzzy variables and F be the objective set. G expresses the input values or named as the raw data,
there are m years to be forecasted (G1, G2, . . . , Gm); Xn is the set of
all schemes. Z is the set of all products values and is decided by
X. So if there are k products in all the schemes, Z can be expressed
as (Z1, Z2, . . . , Zk), Z  Rk, and k may be equal to n or not. Ck is the value of the fuzzy variables, and if there are k products, there must be
k fuzzy variables in the model. Fk are the set of all objectives that
the decision-maker look forward. Actually, we get the optimal value of fuzzy variables Ck through the process of objectives Fk in
order to run the SD system.
We divided these parameters into ve types: initial, level, rate,
fuzzy and objective. G is the initial parameter, Z is the level
parameter, controlled by initial and rate parameter. X is the rate
parameter, C is the set of fuzzy variables, F is the objective parameters.

Fig. 2. SD-FMOP modeling.

11554

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

Thus, the natural coal industry system includes Gm (m = 1, 2, . . . ,


M); Xn (n = 1, 2, . . . , N); Fk (k = 1, 2, . . . , K); zk; Ck (k = 1, 2, . . . , K). This
dynamics model can be described mathematically as:
Level: level parameter is controlled by the raw data Gm
(m = 1, 2, . . . , M) and rate parameter Xn (n = 1, 2, . . . , N).

Z k Z 1 ; Z 2 ; . . . ; Z k f G1 ; G2 ; . . . ; GM ; X 1 ; X 2 ; . . . ; X N

Objective: objective function is affected by level parameters and


fuzzy variables.

F k F 1 ; F 2 ; . . . ; F K f Z 1 ; Z 2 ; . . . ; Z K ; C 1 ; C 2 ; . . . ; C K

Fuzzy: fuzzy variables are the price of products and affected by


many other factors.

C k C 1 ; C 2 ; . . . ; C K lj ; aj ; bj j 1; 2; . . . ; K

The general SD-FMOP modelling may be described mathematically as follows:

8
max F k f G1 ; G2 ; . . . ; GM ; X 1 ; G2 ; . . . ; X N ; C k
>
>
>
>
>
< subject to :
g i X 1 ; X 2 ; . . . ; X N < 0
>
>
>
C k C 1 ; C 2 ; . . . ; C K lj ; aj ; bj
>
>
:
i 1; 2; . . . ; N; j 1; 2; . . . ; K

The model aims to maximize Fk which is affected by the raw


data GM, the scheme sets XN and the fuzzy variables Ck,gi are the
constraints based on the scheme sets XN.
3.2. Solution method
There are many methods that lead to effective solutions to the
problem. We adopt chance-constraint programming (CCP) and genetic algorithm (GA) to solve the problem of uncertain programming. CCP offers a powerful means of modelling uncertain
decision systems under the assumption that the fuzzy constraints
will hold at least a where is referred to as the condence level provided as an appropriate safety margin by the decision-maker.
3.2.1. Chance-constraint programming
This section explains how CCP can be used to transform the fuzzy variable model into a critical variable one.
Denition 3.1. A point x is called feasible if and only if the
probability measure of the event gj(x, n) 6 0, j = 1, 2, . . . , p is at least a.
In other words, the constraints will be violated at most (1  a)
times. Sometimes, the joint chance constraint is separately
considered as

Posfg j x; n 6 0g P bj ;

j 1; 2; . . . ; p

which is referred to as a separate chance constraint.


There are two methods of chance-constrained programming
(CCP) separated as maximax (optimistic) chance-constrained programming and minimax (pessimistic) chance-constrained programming. In this paper, as we need obtain the optimistic levels
to the different values, we adopt the maximax chance-constrained
programming which can obtain feasible optimistic values. In order
to maximize the optimistic return with a given condence level
subject to some chance constraint, Liu applicate the following CCP:

8
>
max f
>
>
>
>
< subject to :
>
Posff x; n P f g P a
>
>
>
>
:
Posfg j x; c 6 0; j 1; 2; . . . ; pg P b

where a and b are the predetermined condence levels. In practice,


we may have multiple objectives. Thus, we have to employ the following chance-constrained multi-objective programming (CCMOP):

8
>
maxf 1 ; f 2 ; . . . ; f m 
>
>
>
< subject to :
>
Posff x; n P f i g P ai ;
>
>
>
: Posfg x; c 6 0g P b ;
j

i 1; 2; . . . ; m
j 1; 2; . . . ; p

where max fi are the ai-optimistic values to the return functions
fi(x, n), i = 1, 2, . . . , m, respectively.
3.2.2. GA
Researchers of various backgrounds have increasing interest on
multiple-objective optimization problems since the early 1960s
(Hwang & Yoon, 1981). Genetic algorithms (GAs) have received
considerable attention regarding their potential as a novel approach to multi-objective optimization problems, resulting in a
fresh body of research and applications known as genetic multiobjective optimizations.
GAs do not need many mathematical requirements and any
type of objective functions and constraints.For many real-world
problems, the set of Pareto solutions may be very large so that it
is hard to solve those. In addition, to evaluate a large set of Pareto
solutions and to select the best one poses a considerable cognitive
burden on DM. Therefore, in this case, obtaining the entire set of
Pareto solutions is of little interest to DMs. To overcome such difculty, Gen and Cheng (2000) proposed a compromise approach
which intends to search for compromised solutions instead of generating all Pareto solutions.
In this section, we will refer to Gen and Cheng (2000) and Li
et al. (2006), and present a fuzzy simulation-based genetic algorithm to obtain a compromise solution of multi-objective programming with fuzzy variable. The detail will be introduced as follows.
Representation. A reasonable representation structure of the
solution is very critical for the genetic algorithm. Model (7) is a
multi-objective programming problem respect to the continuous
decision vector x. Thus a vector x = (x1, x2, . . . , xn) satisfying constraint conditions Posff x; n P f i g P ai and Pos{gj(x, c) 6 0} P bj
is randomly chosen as a chromosome to represent a solution to
the optimization problem. Repeat this process Npopsize times, then
we have Npopsize initial feasible chromosomes x1 ; x2 ; . . . ; xNpopsize .
Evaluation and selection. For the multi-objective programming problem (7), an efcient evaluation indexes should be proposed to get more optimal solutions. The compromise approach,
which is regarded as a kind of mathematical formulation of goalseeking behavior in terms of a distance function, is an efcient
method to obtain the ideal point. In this case, the compromise approach is given as follows: Suppose the ideal point of the DM is
f  f  ; f  ; . . . ; f  T . For each feasible solution x, the regret function
1 2
m
r(f, p)(p P 1) is dened by the following weighted Lp  norm:

rf ; p kf  f  kp;w

"
q
X

#1=p
wpj jf j  f j jp

j1

where weights w = (w1, w2, . . . , wq) are assigned to signal different


degrees of importance. However, for many complex problems, to
nd an ideal point is also a difcult task. Therefore, Gen and Cheng
(2000) introduced the concept of a proxy ideal point to replace the
ideal point. Let P denote the set of the current population, the proxy
max
max
ideal point f max
1 ; f 2 ; . . . ; f q  is calculated as follows:

8

>
>
> max f k
>
< subject to :
>
Posff x; n P f k g P ak
>
>
>
: Posfg x; c 6 0g P b ;
j

8
j 1; 2; . . . ; p

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

11555

all generate new chromosomes from the parents by some probability (see Fig. 3).
The detail of crossover operation is as follows: Generate a random number k from the open interval (0, 1) and the chromosome
xi is selected as a parent provided that k < Pk, where parameter Pk
is the probability of crossover operation. Repeat this process
Npopsize times and Pk  Npopsize chromosomes are expected to be
selected to undergo the crossover operation. The crossover
operator on x1 and x2 will produce two children y1 and y2 as
follows:
Fig. 3. Genetic operator.

y1 kx1 1  kx2 ;

Thus, the evaluation function can be constructed as follows,

ev alx

r max  rx; p e
r max  r min e

where e is a positive real number usually restricted within the open


interval (0, 1).
The selection process is based on spinning the roulette wheel
Npopsize times. Each time a single chromosome for a new population is selected in the following way: Calculate the cumulative
probability qi for each chromosome xi,

q0 0;

qi

i
X

ev alxj ;

i 1; 2; . . . ; Npopsize :

j1

Generate a random number r in 0; qNpopsize  and select the ith chromosome xi such that qi1 < r 6 qi, 1 6 i 6 Npopsize. Repeat the above process Npopsize times and we obtain Npopsize copies of chromosomes.
Genetic operators. As the process of evolution, the genetic
operators include two operation: crossover and mutation. They

y2 kx2 1  kx1

If both children are feasible, then we replace the parents with them,
or else we keep the feasible one if it exists. Repeat the above operation until two feasible children are obtained or a given number of
cycles is nished.
Similar to the crossover process, the chromosome xi is selected
as a parent to undergo the mutation operation provided that random number m < Pm, where parameter Pm as the probability of
mutation operation. Pm  Npopsize are expected to be selected after
repeating the process Npopsize times. Suppose that x1 is chosen as
a parent. Choose a mutation direction d 2 Rn randomly. Replace x
with x + M  d if x + M  d is feasible, otherwise we set M as a random between 0 and M until it is feasible or a given number of cycle
is nished. Here, M is a sufciently large positive number. We illustrate the fuzzy simulation-based genetic algorithm procedure as
follows:
Step 0: Input the parameters Npopsize, Pk and Pm.
Step 1: Initialize Npopsize chromosomes whose feasibility may be
checked by fuzzy simulation.

Fig. 4. The coal industry system.

11556

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

Table 1
Fuzzy variables.
c1
(0.18, 0.05, 0.05)

c2
(0.38, 0.05, 0.05)

c3
(6500, 380, 380)

c4
(17,000, 1060, 1060)

c5
(23,500, 670, 670)

c6
(23,500, 520, 520)

c7
(1200, 150, 150)

Step 2: Update the chromosomes by crossover and mutation operations and fuzzy simulation is used to check the feasibility
of offspring.
Step 3: Compute the tness of each chromosome based on the
regret value.
Step 4: Select the chromosomes by spinning the roulette wheel.
Step 5: Repeat the second to fourth steps for a given number of
cycles.
Step 6: Output the best chromosome as the optimal solution.
4. The coal industry system
In this section, we will describe the coal industry system studied in our paper. The system is described diagrammatically in
Fig. 4. We will dene the parameters used to describe and analyze
the system.

x1
x2
x3
x4
x5
x6

Weight
Weight
Weight
Weight
Weight
Weight

of
of
of
of
of
of

coal in electricity.
coal in dimethyl ether.
dimethyl ether in ethylene.
propylene in ovi-ammonia acid.
ethylene in PVC.
coal stone in electricity.

The FMOP will aim for the most output value of each product in
forming the nal objects. As in this coal industry system, there are
seven objectives from f1 to f7 we will establish seven objective
equations to optimize the parameters. The restrictions gj are the
capacities of each product, g1 to g7. We may input the optimum
values into the system dynamics (SD) model to simulate and forecast the output value of products in future years. This proves that
the SD-FMOP model is the best solution for this coal industry system. The following is a list of the objectives and restrictions
considered:
f1 Output value of brick.
f2 Output value of generating electricity.
f3 Output value of PVC.
f4 Output value of epoxy-hexane.
f5 Output value of crylic acid.
f6 Output value of ovi-ammonia acid.
f7 Output value of compositive oil.
gj The max capacity in regional market of each product
(f = 1, 2, . . . , 7).
4.3. Parameters

4.1. The price of products


Factors that affect the prices of products include technology restraints, market factors and relation between supply and demand,
etc. Mere consideration of these factors based on experience cannot sufce; we must consider the cost of products, the price of demand, the price of competition, and many other factors to produce
an accurate prole. As noted earlier, a fundamental assumption
made in the application of fuzzy set theory is that any object can
be perceived through available information, while it may be insufcient for an exact characterization. As before, we approximate the
set by other sets.
In the coal industry system of this region, there are seven fuzzy
variables expressed by c1 to c7. The fuzzy variables contradistinguish the gas system as following and their values in Table 1.
c1
c2
c3
c4
c5
c6
c7

The
The
The
The
The
The
The

price
price
price
price
price
price
price

of
of
of
of
of
of
of

brick.
electricity.
PVC.
epoxy-hexane.
crylic acid.
ovi-ammonia acid.
compositive oil.

4.2. Weight of distribution


The weight of distribution xn is an important parameter for the
coal industry; its value decides the future direction of product
development. An example: x1 expresses the weight of coal in electricity, x2 expresses the weight of coal in compositive oil and
1  X1  X2 expresses the weight of coal used in the synthesis of
ammonia. That is, if x1 + x2 < 0.5, then the main direction of growth
reects the production of dimethyl ether and vice versa. In this
natural gas industry system, n = 6. x1 to x6 decide the nal developmental emphases of seven products which contradistinguish the
coal system as illustrated below. We attempt to introduce the
Fuzzy multi-objective programming (FMOP) to solve this problem.

We elicited parameter statistics by studying coal chemistry


industry chains and analyzing the ows of processing technique.
There are four basic units of measurement: the output of products using 105T, the output value of products using 105 yuan,
the price of products using yuan/T and the weight of one product
in the other product having no dimension. Of course there are
some specic types of units such as the output of brick using
105 piece, the price of brick using yuan/piece, the price of electricity using yuan/degree and the degree of generating electricity
usage.
The settlements for the rates of substance transformation
and some settled parameters in the system dynamic model are
mainly based on the statistics yearbook. In this region the programming reports on correlative industries and present market circumstances. The settlements were obtained via equilibration, linearity
regression, index smoothness and other related mathematical
models.
4.4. Practise
Previously, we were accustomed to setting parameters based on
subjective experience, which lent to inaccuracy, vague decision
making and bias. Our integrated model of system dynamics and
fuzzy multi-objective programming can reduce experimenter bias,
for better qualitative analysis the rate parameters xn are optimized
via fuzzy multi-objective programming. The rate parameters and
fuzzy variables are both included in this system. The initial and
rate parameters determine the corresponding level parameter.
Every objective parameter is compared to a fuzzy multi-objective
programming model. Therefore, we set up the fuzzy multi-objective programming models according to every objective parameter.
For "fi(i = 1, 2, . . . , 7), we set up fuzzy multi-objective programming (FMOP). The following will be the optimizing formulas for
the natural gas industry system of this region. The multi-objective
programming problem with fuzzy parameters is constructed as
follows,

11557

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

8
max f 1 k1 ~c1 1  x6
>
>
>
>
>
max
f2 ~c2 k21 x1 k22 x6
>
>
>
>
>
>
max
f 3 k3 ~c3 x2 x3 x5
>
>
>
>
>
max
f 4 k4 ~c4 x2 x3 1  x5
>
>
>
>
>
max
f 5 k5 ~c5 x1 x4 x10
>
>
>
>
>
~
>
> max f 6 k6 c6 x2 x4 1  x3
>
>
>
>
> max f 7 k7 ~c7 x1 1  x1  x2
>
>
>
>
< subject to :
x1 x 2 < 1
>
>
>
0
< xi < 1; i 1; 2; . . . ; 7
>
>
>
>
>
>
k
1 1  x6 < b1
>
>
>
>
>
k21 x1 k22 x6 < b2
>
>
>
>k x x x < b
>
3 2 3 5
3
>
>
>
>
> k4 x2 x3 1  x5 < b4
>
>
>
>
>
k5 x1 x4 x10 < b5
>
>
>
>
>
k6 x2 x4 1  x3 < b6
>
>
>
:
k7 x1 1  x1  x2 < b7

It follows that y = k1x(1  x6) = k1(1  x6)(c1  a1L1(x))


6 c1 y
x. Similarly, we can prove
Thus we have Lk1k1x
1 1x6 a1
~1 has
x. Then we know that the fuzzy variable s

1 1x6 c1

Ryk
k1 1x6 b1

the following membership function,

8 

>
6 c 1 y
< L k1k1x
; t 6 k1 1  x6 c1
1 1x6 a1


ls~1 t
>
1 1x6 c 1
: R yk
; t P k1 1  x6 c1
k1 1x6 b1
9

Then it follows that


Posfk1 ~c1 1  x6 g P h1 () mLh1 6 f 1 6 nRh1 where mLh1 k1 1
x6 c1  L1 hk1 1  x6 a1 and nRh1 k1 1  x6 c1  L1 hk1 1  x6 b1 .
Since the object is to maximize f 1 , then we only consider the
constraint, f 1 6 k1 1  x6 c1  L1 h1 k1 1  x6 b = k (1  x )(c 
1

L1(h1)b1).
We can get the other equivalent constraints in the same way.
Then the equivalent model is as follows:

Where ~ci i 1; 2; . . . ; 7 are L-R fuzzy variables with the following


membership function,

8  
>
< L ciat ; t 6 ci; ai > 0
i
 
l~ci t
>
i
: R tc
; t P ci ; bi > 0
b

12

i 1; 2; . . . ; 10

10

ai, bi are positive numbers expressing the left and right spreads of
~ci i 1; 2; . . . ; 6 and reference functions L, R:[0, 1] ? [0, 1] with
L(1) = R(1) = 0 and L(0) = R(0) = 1 are non-increasing continuous
functions.
Since the uncertainty of fuzzy variables ~ci i 1; 2; . . . ; 10, it is
difcult for decision makers to make an accurate decision and then
we have to convert it into a deterministic one. The chance operator
is an efcient tool to deal with it when decision makers just want
to get the maximal objective values at a given condence level. Assume that a condence level di is given for each objective fi, we get
the following crisp programming model,
8
max f 1 ; f 2 ; . . . ; f 7
>
>
>
>
>
>
subject to : Posk1 ~c1 1  x6 P f 1 P h1
>
>
>
>

>
~
>
> Posc2 k21 x1 k22 x6 P f 2 P h2
>
>

>
>
> Posk3 ~c3 x2 x3 x5 P f 3 P h3
>
>
>
>
Posk4 ~c4 x2 x3 1  x5 P h4
>
>
>
>
>
> Posk5 ~c5 x1 x4 x10 P h5
>
>
>
>
> Posk6 ~c6 x2 x4 1  x3 P f 6 P h6
>
>
>
>

>
>
< Posk7 ~c7 x1 1  x1  x2 P f 7 P h7
x1 x2 < 1
>
>
>
>
0 < xi < 1; i 1; 2; . . . ; 7
>
>
>
>
>
k
>
1 1  x6 < b1
>
>
>
>
>
k21 x1 k22 x6 < b2
>
>
>
>
>
k3 x2 x3 x5 < b3
>
>
>
>
>
>
k4 x2 x3 1  x5 < b4
>
>
>
>
>
k5 x1 x4 x10 < b5
>
>
>
>
>
>
k6 x2 x4 1  x3 < b6
>
>
:
k7 x1 1  x1  x2 < b7

8
>
max f 1 ; f 2 ; . . . ; f 7
>
>
>
>
>
>
subject to :
>
>
>
>
> f 6 k 1  x c  L1 h b
>
1
1
6
1
1
>
1
>
>
>
>
1
>
f
6
k
x

k
x
c

L
h
>
2
21 1
22 6
2
2 b2 
>
>
>
>
1

>
>
> f 3 6 k3 x2 x3 x5 c3  L h3 b3 
>
>
>
>
> f 4 6 k4 x2 x3 1  x5 c4  L1 h4 b4 
>
>
>
>
>
> f 5 6 k5 x1 x4 x10 c5  L1 h5 b5 
>
>
>
>
>
1
>
>
< f 6 6 k6 x2 x4 1  x3 c6  L h6 b6 
f 7 6 k7 x1 1  x1  x2 c7  L1 h7 b x1 x2 < 1
7
>
>
>
>
>
>
<
1;
i

1;
2;
.
.
.
;
7
0
<
x
i
>
>
>
>
>
>
1

x6 < b 1
k
1
>
>
>
>
>
>
k21 x1 k22 x6 < b2
>
>
>
>
>
>
k3 x2 x3 x5 < b3
>
>
>
>
>
>
k4 x2 x3 1  x5 < b4
>
>
>
>
>
k5 x1 x4 x10 < b5
>
>
>
>
>
>
k6 x2 x4 1  x3 < b6
>
>
>
>
:
k7 x1 1  x1  x2 < b7

13

Since decision makers objective is to maximize f i i 1; 2; . . . ; 10,


the above model is also equivalent to:

11

We take Posk1 ~c1 1  x6 P f 1 P h1 as an example to show that how


we convert it into an equivalent constraint.
~1 k1 ~c1 1  x6 . Since ~c1 is an L-R fuzzy variable and
Denote s
k1 > 0, 0 < x1 < 1, it follows that k1 ~c1 is also an L-R fuzzy variable.
In fact, assume that x 2 [0, 1], let Lc1ax
x, then x = c1  a1L1(x).
1

8
max H1 x k1 1  x6 c1  L1 h1 b1
>
>
>
>
>
>
max H2 x k21 x1 k22 x6 c2  L1 h2 b2 
>
>
>
>
>
>
max H3 x k3 x2 x3 x5 c3  L1 h3 b3 
>
>
>
>
>
>
max H4 x k4 x2 x3 1  x5 c4  L1 h4 b4 
>
>
>
>
>
max H5 x k5 x1 x4 x10 c5  L1 h5 b5 
>
>
>
>
>
>
max H6 x k6 x2 x4 1  x3 c6  L1 h6 b6 
>
>
>
>
>
>
max H7 x k7 x1 1  x1  x2 c7  L1 h7 b7 
>
>
<
x1 x 2 < 1
>
> 0 < xi < 1; i 1; 2; . . . ; 7
>
>
>
>
> f x k 1  x < b
>
1
1
6
1
>
>
>
> f x k x k x < b
>
>
2
21
1
22
6
2
>
>
>
>
>
f
x

k
x
x
x
<
b
3
3
2
3
5
3
>
>
>
>
>
f4 x k4 x2 x3 1  x5 < b4
>
>
>
>
>
f5 x k5 x1 x4 x10 < b5
>
>
>
>
>
>
f6 x k6 x2 x4 1  x3 < b6
>
>
:
f7 x k7 x1 1  x1  x2 < b7

14

11558

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

Table 2
Results.
x1

x2

x3

x4

x5

x6

0.596

0.404

0.371

0.503

0.44

0.94

f1
625

f2
511,115

f3
25,028

f4
45,856

f5
95,858

f6
96,995

^
x1
0.47

^
x2
0.25

^
x3
0.6

^
x4
0.21

^
x5
0.44

^
x6
0.5

^f
1
1704

^f
2
201,341

^f
3
24,924

^f
4
45,747

^f
5
59,852

^f
6
15,910

f7
0

^f
7
29,864

Fig. 6. Output value of generating electricity.

Fig. 5. Output of crylic acid.

5. Numerical results
The numerical results is shown in this section.
Fig. 7. Output of ovi-ammonia acid.

5.1. Optimization and simulation


According to the optimizing formulas above and given solution
methods we can obtain the optimum results xi and fi and compare
them to the current scheme which is marked as ^
xi and ^f i shown in
Table 2.
We then inserted the parameter values obtained from the multi-objective programming model into the system dynamics model.
A system simulation was then performed using the simulation
software VENSIM (Yeh, Wang, & Yu, 2006) and marked the data
from 2006 as the initial conditions, time = 0. Time is measured in
years. Our simulation spans fteen years, i.e., t ranges from one
to fteen, and results in data analysis for Year 2006 to Year 2020.
We depict the main pattern from Fig. 58.
The gures exhibit the ndings obtained from the coal industry
system studied.
(1) As a result of scientic optimization techniques, the coal
industry experienced a sizeable increase in the economic
prots. The main goal of this paper is to increase the total
value of the coal industry by optimizing and adjusting its
developmental structure.
(2) The output value of some products obviously increases. Figures exhibit a signicant increase in those products over the
fteen year period. Therefore, we advise the establishment
of strong collaborations and relations between relevant
industries in this region.
(3) After optimization, we have concluded an emphasis on
increasing importance of the electric power industry. Generating electricity will consume too many coal stones. The
environment will be effectively improved.

Fig. 8. The total value.

(4) Using optimization techniques and simulation forecasting,


we can accurately estimate the direction of the coal industrys development in this region. The results of this study
will assist administrations to establish policies that can
effectively promote further growth and development.
5.2. Sensitivity analysis
Risk analysis of failing to satisfy the optimal objective in the
coal production System is demonstrated through analyzing parameter sensitivity. Risk analysis for Other subsystems can be done in
the same way. In practice, an optimal value for fi computed in the

J. Xu, X. Li / Expert Systems with Applications 38 (2011) 1155211559

Model (8) can hardly be achieved. The central planner usually allows a variation from the optimal objective, which motivates the
utilization of a coefcient a(0 < a < 1) in the objective functions.
Therefore, we consider the result is rational and acceptable if the
actual value of the objective is within the range (a fi, fi ), which
can be explained as an accepted or allowable risk interval by the
decision-maker.
We examine the effect of changing these variables to the value
of objective functions by repeating the simulation. The simulation
shows altering x5 in the range of 10% makes the value of associated
objective functions change up to 612%, while changing in other
variables almost has no effect on the objective functions. For example, given a = 0.85, we get x5 = 0.484, which means the output value of PVC f3 or the output value of epoxy-hexane f4 can reach at
least 85% of its maximum if the decision-maker limits the sensitive
variable x5 (other decrement) in the interval (0.44, 0.484). We then
communicate with the decision-makers. If they are satised with
this result, we put this result into the SD model in order to do
the simulation. If they are not satised with the result, we should
adjust the constraints and do more calculation to obtain a preferred result.
6. Conclusion
In this paper, we have developed a model that integrates system
dynamics with fuzzy multi-objective programming (SD-FMOP).
This model can be used to study the complex interactions in a coal
industry system. In the process of conrming sensitive parameters
and fuzzy variables of the SD model, we made use of fuzzy multiobjective programming to help yield the solution. In this process,
we adopted the chance-constraint programming model to convert
the fuzzy variables into precise values. We use an genetic
algorithm to solve FMOP model, and obtain the Pareto solution
through the programming models. The results recorded from the
SD model are in our option, reasonable and credible. These results
may help governments to establish more effective policy related to
the coal development.
Acknowledgment
This research has been supported by the Key Program of NSFC
(Grant No. 70831005) and the National Science Foundation for Distinguished Young Scholars, PR China (Grant No. 70425005).

11559

References
Aurora, R., Rodney, A., Rolando, G., Jack, G., & Thomas, R. (2005). Adsorption of Hg
and NOx on coal by-products. Fuel, 84, 911916.
Ball, A., Hansard, A., Curtotti, R., & Schneider, K. (2003). Chinas changing coal
industryimplications and outlook,Canberra: ABARE eReport. Available from:
<http://abareonlineshop.com/product.asp?prodid=12494i>.
Chakraborty, M., & Chandra, M. K. (2005). Multicriteria decision making for optimal
blending for beneciation of coal: A fuzzy programming approach. Omega, 33,
413418.
Chen, Y., Qi, J., et al. (2004). Dynamic modeling of a man-land system in response to
environmental catastrophe. Human and Ecological Risk Assessment, 10, 579593.
Fan, Y., Yang, R.-G., & Wei, Y.-M. (2007). A system dynamics based model for coal
investment. Energy, 32, 898905.
Gen, M., & Cheng, R. (2000). Genetic algorithms and engineering optimization. New
York: Wiley.
Grzegorz, S. (2006). Some technical issues of zero-emission coal technology.
International Journal of Hydrogen Energy, 31, 10911102.
Hwang, C., & Yoon, K. (1981). Multiple attribute decision making: Methods and
applications. Berlin: Springer-Verlage.
Jaffrennou, C. et al. (2007). Simulations of accidental coal immersion. Marine
Pollution Bulletin. doi:10.1016/j.marpolbul.
Jay, W., Nathaniel, J., Mass Charles, J., & Ryan (1976). The system dynamics national
model: Understanding socio-economic behavior and policy alternatives.
Technological Forecasting and Social Change, 9, 185192.
Jorjani, E., Chelgani, S., & Mesroghli, S. (2007). Prediction of microbial
desulfurization of coal using articial neural networks. Minerals Engineering, 20,
12851292.
Li, J., Xu, J., & Gen, M. (2006). A class of multiobjective linear programming model
with fuzzy random coefcients. Mathematical and Computer Modelling, 44,
10971113.
Min, K., Kyung, J., & Moosung (2005). A quantitative assessment of LCOs for
operations using system dynamics. Reliability Engineering System Safety, 87,
211222.
Naill, R., & Geinzer, G. (1993). Integrated dynamic energy analysis simulation.
Virginia: The AES Corporation.
Noriaki, E. (2000). R&D of coal utilization technology in Japan. Fuel Processing
Technology, 62, 143151.
Pendharkar, P. C. (1997). A fuzzy linear programming model for production
planning in coal mines. Computers and Operations Research, 24(12), 11411149.
Shih, J.-S., & Christopher Frey, H. (1995). Coal blending optimization under
uncertainty. European Journal of Operational Research, 83, 452465.
Suwala, W. (2008). Modelling adaptation of the coal industry to sustainability
conditions. Energy, 33, 10151026.
Suwala, W., & Labys, W. C. (2002). Modelling adaptation of the coal industry to
sustainability conditions. Energy Economics, 24, 285303.
Wigley, F., Williamson, J., & Riley, G. (2007). The effect of mineral additions on coal
ash deposition. Fuel Processing Technology, 88, 10101016.
Yeh, S., Wang, C. A., & Yu, H. C. (2006). Simulation of soil erosion and nutrient
impact using an integrated system dynamics model in a watershed in Taiwan.
Environmental Modelling and Software, 21, 937948.
Zhang, H., & Zhao, G. (1999). CMEOC An expert system in the coal mining industry.
Expert Systems with Applications, 16, 7377.

You might also like